Systematic Review on Predictive Utility of Electrocardiographic Correlates for Cardiovascular Disease

Cardiovascular disease (CVD) is the leading cause of global mortality. Electrocardiography (ECG) is widely accessible and non-invasive, and its abnormalities may carry prognostic value beyond diagnosis. Artificial intelligence (AI) has expanded ECG's risk-prediction potential, though its incremental value remains debated. This systematic review evaluated the predictive utility of conventional (non-AI) and AI/deep learning-enhanced ECG approaches for cardiovascular disease incidence, mortality, and major adverse cardiovascular events, comparing the two directly. The review followed PRISMA 2020 guidelines. Studies were identified through PubMed/MEDLINE, Scopus, Web of Science, and Google Scholar. Eligible studies were cohort studies, systematic reviews, or validated AI-based ECG investigations reporting risk estimates or predictive performance metrics. Findings were synthesized narratively and tabulated descriptively rather than statistically pooled. Thirty core studies were included across general, elderly, hypertensive, diabetic, and coronary artery disease cohorts. Conventional ECG abnormalities QT/QTc prolongation, QRS duration, left ventricular hypertrophy, ST-T abnormalities, and atrial fibrillation were consistently associated with cardiovascular disease, mortality, and MACE (hazard ratios generally 1.2–2.5). Adding ECG variables to conventional models yielded modest gains (ΔAUC ≈ 0.02–0.06); AI/deep learning models outperformed conventional interpretation (ΔAUC ≈ 0.10–0.15), including via ECG derived biological age. Conventional and AI-enhanced ECG correlates both carry meaningful, independent prognostic information.